Skip to content
chenkb98Public

About

Reproducible audio generation experiments with synthesis and diffusion baselines, manifests, and metrics.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

sonoforge

CI

Reproducible audio generation experiments with synthesis and diffusion baselines, manifests, and metrics.

Chen Kaibin · 华南理工大学在读。项目关注音频生成和语音大模型研究中的可复现工具。

本仓库的 2025–2026 提交时间线为重建的演示数据,不代表实际开发日期。

快速开始

python -m venv .venv
.venv/bin/python -m pip install -r requirements-dev.txt

激活虚拟环境后运行:

sonoforge generate --prompt "sine 440" --duration 0.1 --output outputs/tone

主要功能

扩散模块提供可训练的仿射噪声预测基线;实验模块统一合成与采样接口。短窗口扩散不会自动保证跨窗口连续性。

内容 入口
使用与参数 使用文档
完整模块 API API 索引
数据流与边界 架构
数值定义与参考 设计说明
可运行合成示例 examples

验证

make build
make test
make format-check
make lint typecheck

CI 覆盖 Python 3.11 和 3.12。开发工具版本由 requirements-dev.txt 固定。 测试包含空输入、无效值、往返转换、完整字段模式和公共接口契约。

范围

默认流程不需要 GPU、付费服务、模型权重或外部数据。音频示例与训练示例是合成基线, 不能用来宣称真实语音准确率或感知质量。PCM16 会量化并饱和削波;线性重采样没有抗混叠滤波。

MIT 许可。参与方式见 CONTRIBUTING。

可选模型接口

见上游模型适配器。默认验证不加载模型权重。

About

Reproducible audio generation experiments with synthesis and diffusion baselines, manifests, and metrics.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages